Dominance Analysis

Dominance Analysis: Concept Architecture

Orientation and learning roadmap

The page first establishes a common basis for comparing the available interventions. It then shows how to remove clearly inferior choices, test the remaining options for extended dominance, and calculate incremental ratios only along the resulting frontier. A worked example and final interpretation section show what the screening does, and what uncertainty or decision context it cannot settle.

Put the alternatives on the same footing

Dominance comparisons require every strategy to be assessed for the same population, perspective, time horizon, outcome measure, and cost basis. Costs and effects should be expected values for each mutually exclusive option, including the relevant current practice; changing any of these assumptions can change which option appears inferior. For health outcomes such as quality-adjusted life-years (QALYs), more is treated as better, while lower cost is preferred.

Identify strict dominance

An alternative is strictly dominated when another feasible alternative costs no more and produces at least as much benefit, with at least one strict improvement. Thus a more expensive, less effective treatment is unambiguously inferior on these two measured dimensions in the specified analysis. Equality on one dimension still permits strict dominance if the other is strictly better; identical cost and effect pairs are ties rather than strict dominance.

For strategies $i$ and $j$, with expected cost $C$ and expected effect $E$, strategy $i$ is strictly dominated by $j$ if

$$C_j\leq C_i,\qquad E_j\geq E_i,$$

with at least one strict inequality. Screening should compare each option against all feasible alternatives, not only the next row in a cost-sorted table. A negative incremental cost-effectiveness ratio (ICER) alone cannot identify dominance because it also occurs when an intervention is cheaper and more effective.

Test for extended dominance and construct the frontier

After strictly dominated choices are removed, sort the remaining options by increasing cost and calculate incremental cost per additional unit of effect. An intermediate option is extendedly dominated when moving to it from a cheaper option has a higher ICER than moving onward to a more effective, more costly option. Equivalently, its cost-effect pair lies above the efficient line segment joining other relevant options when effect is on the horizontal axis and cost on the vertical axis.

For adjacent surviving strategies $a$ and $b$ with $E_b>E_a$, the incremental ratio is

$$\operatorname{ICER}_{a\to b}=\frac{C_b-C_a}{E_b-E_a}.$$

Remove an extendedly dominated intermediate option, recalculate ratios against the newly adjacent option, and repeat until ICERs along the frontier rise with effect. The line-segment interpretation presumes that a population-level mix or divisible scale is meaningful; for an individual patient, a hypothetical mixture of indivisible strategies may not itself be a feasible treatment. It still identifies an intermediate option that cannot maximise expected net monetary benefit for any positive willingness-to-pay threshold in the specified two-dimensional analysis.

Work through four strategies

Consider four mutually exclusive strategies evaluated over the same horizon, with costs in pounds per patient and effects in QALYs per patient. The values are illustrative expected outcomes, not observed evidence for a real technology. Compare A and B first, then check whether C sits on the frontier connecting A and D.

B costs £2,000 more than A while yielding 0.1 fewer QALYs, so B is strictly dominated. After removing B, the A-to-C ICER is £4,000/0.2 = £20,000 per QALY, whereas the C-to-D ICER is £2,000/0.3 ≈ £6,667 per QALY. Since the second increment buys health more cheaply than the first, C is extendedly dominated; the remaining A-to-D ICER is £6,000/0.5 = £12,000 per QALY.

At 5.2 QALYs, a population mixture that allocates 60% to A and 40% to D has expected cost £12,400, below C's £14,000. The mixture demonstrates the geometric frontier comparison, subject to its feasibility assumptions. Even without implementing that mix, C does not have the highest expected net monetary benefit at any nonnegative threshold for these inputs.

Interpret the result in a decision

The frontier narrows the set of potentially efficient strategies; it does not itself select one without a decision criterion. If a decision maker uses a willingness-to-pay threshold $\lambda$ per QALY, compare expected net monetary benefit $\lambda E_i-C_i$ across feasible strategies. In the example, A is preferred below £12,000 per QALY, D above £12,000, and they tie at £12,000, provided the stated values and scope hold.

Dominance is conditional on the costs, outcomes, comparators, and model assumptions supplied. Parameter uncertainty can change which strategy is dominated in a simulation, so probabilistic sensitivity analysis should examine the joint distribution of cost and effect and the probability of optimality across relevant thresholds. Equity, budget constraints, clinical suitability, and feasibility may also affect a real decision even when a strategy is excluded from a two-dimensional cost-effectiveness frontier.

Avoid common reporting errors

Dominance analysis is most useful when it is fully incremental and transparent about what was removed at each step. Report the original strategy set, cost and effect estimates, reasons for exclusion, and ratios recalculated against the next surviving comparator. Keep the screen distinct from claims of statistical certainty or from other uses of “dominance analysis” in multi-criteria decision analysis.

• Do not compare each strategy only with the same baseline and then call every resulting ratio an incremental frontier ICER.
• Do not interpret a negative ICER without separately checking the directions of the cost and effect differences.
• Do not call a strategy extendedly dominated merely because its ICER exceeds an external threshold; extended dominance compares it with other feasible strategies.
• Do not omit an important comparator, since adding one can change the frontier and every relevant incremental ratio.
• Do not treat deterministic dominance as proof that an option remains dominated under parameter, structural, or population uncertainty.

Sources and technical basis

The distinction between strict and extended dominance, and the expectation of a fully incremental presentation, follow NICE's methods guidance. The frontier and expected net benefit connection are supported by methodological research on identifying cost-effectiveness frontiers. These sources support the method; the numerical example above was constructed for this page and is independently checkable.

• NICE, Technology appraisal and highly specialised technologies guidance: the manual, economic evaluation.
• NICE, Methods for the development of NICE public health guidance, economic-evaluation quality appraisal checklist.
• NICE, Guide to the methods of technology appraisal 2013, glossary.
• An efficient, non-iterative method of identifying the cost-effectiveness frontier, methodological article.